AI in Commercial Cleaning Only Works on the Right Workflows. Here's How to Pick Them.

AI in commercial cleaning is not something you install. It is something you point at one job. Pointed at the right job, it takes hours of office work off an owner every week. Pointed at the wrong one, it is an expensive way to create more work. The whole skill is telling the two apart.

That idea is not ours. Zac Wolf, senior vice president of product at the facilities software company ServiceChannel, made it in a piece for Buildings magazine in September, AI in facilities management: what works, what doesn't, and where to start. His argument, in short: AI pays off when it is added to an existing workflow at the point where that workflow keeps breaking, with people still making the judgment calls, and on data that is clean. His examples are repairs and work orders. This article applies the same thinking to a cleaning company.

Key takeaways

• JLL's October 2026 survey of 1,321 organizations found 45% already scaling or optimizing AI in facilities management, expected to reach 86% by 2030.
• The same survey names data readiness as one of the biggest barriers to scaling it.
• Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027.
• A workflow is right for AI when it passes five tests: it is frequent, rule-based, costly when it breaks, has its data in one place, and can be checked by a person.
• Start with one workflow, measure it for two weeks, run it for 90 days, then add the next.

Why does AI work in some places and fail in others?

Because most of the effort goes into the wrong jobs.

The numbers from the wider economy say so plainly. In McKinsey's 2025 survey on the state of AI, 39% of respondents said AI had any effect on their company's profit at all, and most of those put it under 5%. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear value or weak risk controls.

Facilities teams are not holding back. JLL's October 2026 report, based on 1,321 organizations, found 45% already scaling or optimizing AI in facilities management and expects that to reach 86% by 2030. It also names the thing that stops them: data readiness is one of the biggest barriers to scaling it.

Put those together and the pattern is clear. The technology is not the problem. Choosing the job, and having the information it needs, is.

What makes a workflow right for AI?

Five tests. A workflow worth handing to AI passes most of them.

1. It happens often. Every night, or every week. A job that happens twice a year is not worth automating, however annoying it is.
2. It follows clear rules most of the time. "Call the people trained for this building, closest first, who are not already at 40 hours" is a rule. "Decide whether this client is worth keeping" is not.
3. It costs you when it breaks. A missed shift, a quote that goes out on Friday instead of Tuesday, an invoice nobody chases. If nothing bad happens when it slips, there is nothing to gain.
4. The information it needs is already in one place. The crew list, the schedule, the rates, the scope, the client's contact. If it lives in your head and three group chats, AI cannot act on it, and neither can a new supervisor.
5. A person can check it before it matters. You approve the quote before it is sent. You see who took the shift. When something looks wrong, it stops and asks.

This is what Wolf means by keeping judgment with people. The AI does the repetitive part. A person still owns the decision.

Which cleaning workflows pass the tests?

Here are the common office jobs in a cleaning company, scored against the five.

Workflow	Often	Clear rules	Costly when it breaks	Data in one place	Checkable	Verdict
Covering a call-off	Nightly	Yes	Yes, the client notices	If the crew list is current	Yes	Start here
Following up a quote	Weekly	Yes	Yes, the bid goes cold	If the bid is saved	Yes	Start here
Chasing a late invoice	Monthly, per client	Yes	Yes, cash	Yes, in your invoicing	Yes	Good
Turning an inspection into a client report	Monthly	Mostly	Yes, at renewal	If inspections are recorded	Yes	Good
Pricing a new building from scratch	Occasional	Partly	Yes	Rarely	Yes	Helper only
Deciding to drop a client	Rare	No	Yes	No	Yes	Keep it human
Handling an angry property manager	Unpredictable	No	Yes	No	Not in time	Keep it human

The first four have something in common: they are the work that follows an owner home. They are not hard, they are relentless. That is exactly where AI earns its keep. We wrote about the first one in detail in building a callout chain that covers a night shift, and it is worth having that chain on paper before any software runs it.

Which jobs should stay with a person?

The ones that depend on judgment, relationships or information nobody has written down.

Pricing a new building is a good example of where AI should help without deciding. It can do the arithmetic: production rates, loaded labor, supplies, margin, the way how to bid a commercial cleaning contract lays it out. But the walkthrough, the read on the property manager and the final number are the owner's. Treat it as a calculator that never gets tired, not as the person who signs.

A property manager who calls upset is the clearest case of all. That call is worth more than any report you will ever send, and it should go to the person who owns the relationship.

Where should a cleaning company start?

With one workflow, not a strategy. Wolf's advice translates directly:

1. Pick the job that scores best. For most companies that is covering call-offs or following up quotes.
2. Measure it for two weeks before anything changes. How long from call-off to shift covered, in minutes. How long from walkthrough to quote sent, in days. Most owners have never timed either.
3. Get the data in one place first. Current crew list with phone numbers and languages, every client and job, every rate. This is the step everyone wants to skip, and it is the one JLL's survey says holds people back.
4. Run it for 90 days. Long enough to see a slow week and a bad week.
5. Compare against your two weeks. If the minutes and days went down, add the next workflow. If not, you have learned something cheap.

What should property managers ask a contractor who says they use AI?

More and more cleaning contractors will say it this year. Four questions tell you whether it means anything:

1. Which job does it do? A specific answer ("it covers call-offs") is a good sign. "It helps with everything" is not.
2. How did you measure it? Before and after, in minutes or days.
3. Who checks the result before it reaches me? Every report and every message should pass a person.
4. What happens when it gets something wrong? There should be a plain answer, and a person accountable.

The contractors who answer those four easily are the ones using it on the right workflows.

taskforse is building AI for exactly those first four workflows, with an owner approving what matters. If you run a cleaning company and still cover the shift and send the quote at night yourself, join the waitlist.